Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Let B∈Mn(Z)B\in M_n(\mathbb Z) be the nonsingular upper-triangular matrix from [[additive_combinatorics/adamczewski_2026_erdos1/lattice_reduction|the lattice reduction]]. Let SS be the (n+1)×(n+1)(n+1)\times(n+1) matrix whose first column and last row vanish and whose block in the first nn rows and last nn columns is BB. Because BB is upper triangular, every nonzero entry of SS lies strictly above the diagonal. For an integer tt, put

Ut=I+tS.U_t=I+tS.

This is an integer upper-unitriangular matrix with determinant 11.

Define the unimodular balancing map

β(x0,…,xn)=(x0,…,xn−1,xn−x0−⋯−xn−1)\beta(x_0,\ldots,x_n) =(x_0,\ldots,x_{n-1},x_n-x_0-\cdots-x_{n-1})

and, for z∈Znz\in\mathbb Z^n,

Φt(z)=βUt(0,z).\Phi_t(z)=\beta U_t(0,z).

Direct multiplication gives

Φt(z)=tLB(z)+E(z),LB(z)=(Bz,−∑i(Bz)i),E(z)=β(0,z).(1)\Phi_t(z)=tL_B(z)+E(z),\qquad L_B(z)=\left(Bz,-\sum_i(Bz)_i\right),\quad E(z)=\beta(0,z). \tag{1}

Put

A(B)=∑i,j∣adj⁡(B)ij∣,EB=(n+1)A(B).A(B)=\sum_{i,j}|\operatorname{adj}(B)_{ij}|,\qquad E_B=(n+1)A(B).

Statement

The error term is dominated by the balanced norm: each z∈Znz\in\mathbb Z^n satisfies

∥E(z)∥∞≤EB∥LB(z)∥∞.\|E(z)\|_\infty\leq E_B\|L_B(z)\|_\infty.

Proof

Put M=∥LB(z)∥∞M=\|L_B(z)\|_\infty, so that ∥Bz∥∞≤M\|Bz\|_\infty\leq M by the definition of LBL_B. The integer det⁡B\det B is nonzero, so ∣det⁡B∣≥1|\det B|\geq1, and

(det⁡B)z=adj⁡(B)Bz.(\det B)z=\operatorname{adj}(B)Bz.

Consequently every ∣zi∣≤A(B)M|z_i|\leq A(B)M. The first nn coordinates of E(z)=β(0,z)E(z)=\beta(0,z) are drawn from 0,z1,…,zn−10,z_1,\ldots,z_{n-1}, while its last coordinate is zn−z1−⋯−zn−1z_n-z_1-\cdots-z_{n-1}. Each is bounded in absolute value by (n+1)A(B)M=EBM(n+1)A(B)M=E_BM.

Source and dependencies

An explanation of the proof of Erdős Problem 1, preliminary exposition with no named author (erdosproblems.com, 2026), §5, equations (17)–(19) and Lemma 5.1, pp. 6–7. The edition read is named on the source card. The matrix identities are exact; the estimate uses only the adjugate identity.

Bears on. #1.